A composition-dependent molecular clutch between T cell signaling condensates and actin

During T cell activation, biomolecular condensates form at the immunological synapse (IS) through multivalency-driven phase separation of LAT, Grb2, Sos1, SLP-76, Nck, and WASP. These condensates move radially at the IS, traversing successive radially-oriented and concentric actin networks. To understand this movement, we biochemically reconstituted LAT condensates with actomyosin filaments. We found that basic regions of Nck and N-WASP/WASP promote association and co-movement of LAT condensates with actin, indicating conversion of weak individual affinities to high collective affinity upon phase separation. Condensates lacking these components were propelled differently, without strong actin adhesion. In cells, LAT condensates lost Nck as radial actin transitioned to the concentric network, and engineered condensates constitutively binding actin moved aberrantly. Our data show that Nck and WASP form a clutch between LAT condensates and actin in vitro and suggest that compositional changes may enable condensate movement by distinct actin networks in different regions of the IS.


Introduction
Biomolecular condensates are compartments in eukaryotic cells that concentrate macromolecules without an encapsulating membrane (Banani et al., 2017;Shin and Brangwynne, 2017). Numerous condensates are found in the cytoplasm and nucleoplasm, where they are involved in processes ranging from mRNA storage and degradation to DNA repair and ribosome biogenesis (Brangwynne et al., 2011;Feric et al., 2016;Luo et al., 2018;Protter and Parker, 2016). They are also found at membranes, where they control the organization, and likely the activity, of many signaling receptors (Banjade and Rosen, 2014;Su et al., 2016;Zeng et al., 2018).
Condensates are thought to form through phase separation driven by multivalent interactions between molecules containing multiple binding elements (Banani et al., 2017;Banjade and Rosen, 2014;Li et al., 2012). Recent models suggest that a limited collection of proteins and/or RNA molecules forms the essential phase separating scaffold of particular condensates (Banani et al., 2016;Ditlev et al., 2018;Langdon et al., 2018). These molecules then recruit larger numbers of client proteins to complete the structure. Condensate composition is thus determined by the specificity of interactions among scaffolds and between scaffolds and clients.
The concentrations, interactions, and dynamics of scaffolds and clients are believed to dictate the biochemical activities, and consequent cellular functions, of individual condensates (Banani et al., 2017;Holehouse and Pappu, 2018;Stroberg and Schnell, 2018). The composition of many condensates is known to change in response to signals, implying regulated changes in activity (Chen et al., 2008;Dellaire et al., 2006;Markmiller et al., 2018;Salsman et al., 2017;Youn et al., 2018). However, the relationships between composition and biochemical/cellular functions are not well understood in most cases. 5 switching between compositions and actin-binding modes enables them to move radially via the two actin networks at the IS.

LAT condensates reconstituted in vitro within actin networks move in a compositiondependent manner
We previously reconstituted LAT condensates on supported lipid bilayers (SLBs) through addition of various T cell signaling proteins to membrane-attached phospho-LAT (pLAT) (Su et al., 2016).
In separate work, we also reconstituted membrane-associated contractile actomyosin networks by attaching actin to SLBs via the membrane-anchored actin binding domain of ezrin (eABD) in the presence of myosin II and capping protein (Köster et al., 2016).
To examine interactions of LAT condensates with actin networks, here we combined these two systems into a single assay ( Figure 1A). We attached polyhistidine-tagged phospho-LAT (pLAT) and polyhistidine-tagged eABD to Ni-NTA functionalized lipids within the SLB. We induced LAT phase separation into condensates by adding an increasing subset of binding partners, in the order Grb2, Sos1, phospho-SLP-76 (pSLP-76), Nck, and, finally, N-WASP, as previously described (Su et al., 2016). Hereafter we use the nomenclature pLAT à X to indicate condensates containing pLAT and all binding partners up to X (e.g. if X is Nck, then the condensates would contain pLAT, Grb2, Sos1, pSLP-76, and Nck). Note that in T cells, the main WASP family protein at the IS is WASP (Kumari et al., 2014), which acts as a constitutive complex with WASP Interacting Protein (WIP) (Anton et al., 2002;Ramesh et al., 1997).
However, due to difficulties expressing recombinant full length WASP, here we used full length N-WASP fused N-terminally to the N-WASP binding fragment (residues 451-485) of WIP.
WASP and N-WASP are 48 % identical in overall amino acid sequence, and 56 % identical in the regions that appear to be most important in the biochemical context of phase separation (Basic and Proline-rich, see below and Su et al. (Su et al., 2016)). Further, the known binding interactions and regulation of WASP and N-WASP are highly analogous (Padrick and Rosen, 2010). Thus, while different in an organismic context (Jain and Thanabalu, 2015;Snapper et al., 2001Snapper et al., , 1998, the essential biochemical behaviors of WASP are very likely to be reflected in N-WASP. We fused N-WASP to the N-WASP binding fragment of WIP in order to stabilize the EVH1 domain of the protein (Peterson et al., 2007;Volkman et al., 2002), to create a single, well-expressed polypeptide. Our use of full length N-WASP here represents a step closer to the natural signaling system than our previous reconstitution, which used only a C-terminal fragment, enabling us to better capture essential features of cellular LAT condensates while still maintaining manageable complexity.
For experiments involving actin, we added polymerized actin filaments that bound to the SLB via anchored eABD. To induce actin filament movement we added muscle myosin II and ATP, as previously described (Köster et al., 2016). While T cells express only non-muscle myosin II, the muscle isoform is functionally similar, differing largely in making somewhat longer filaments (800 nm vs 300 nm average length under similar conditions (Vicente-Manzanares et al., 2009)), and is much easier to purify from tissues in biochemical quantities. Since most movement of T cell receptor condensates (which typically coincide with LAT condensates) is blocked by inhibition of myosin (Yi et al., 2012), more complex reconstitutions including actin filament assembly and disassembly dynamics were not warranted in our work here (Blanchoin et al., 2000;Didry et al., 1998;Shekhar and Carlier, 2017). Thus, our reconstituted system should retain key qualitative behaviors of T cell actomyosin, while remaining experimentally practical.
We induced LAT condensate formation without actin, with actin alone, or with active actomyosin networks ( Figure 1B, Videos 1, 2, and 3). We immediately observed that condensates containing Nck or Nck and N-WASP associated with and wet actin filaments in both actin networks alone and actomyosin networks, while those lacking these proteins remained distributed across the SLB ( Figure 1B). As a corollary, co-localization analysis (see Supplemental Methods) showed that actin enrichment in condensates increased significantly in the presence of Nck and N-WASP ( Figure 1C).
In all conditions, we automatically detected and tracked the condensates at the SLB for 15 minutes, and then classified their movement on the SLB using Moment Scaling Spectrum analysis ( (Jaqaman et al., 2011(Jaqaman et al., , 2008Vega et al., 2018); see Materials and Methods) ( Figure 1 -figure supplement 1). This analysis revealed that LAT condensates were either immobile, confined (i.e. moved within a confinement region), or mobile (i.e. moved without apparent restrictions, in a manner akin to free diffusion). In the absence of actin, 80% of condensates showed little movement, regardless of composition; they were either immobile or confined, in an area of radius 100-150 nm ( Figure 1D, E, Video 1). In the presence of actin alone (i.e. no myosin), condensate movement varied with composition. pLAT à Sos1 and pLAT à pSLP76 showed an increase in apparent diffusion coefficient, the mobile fraction, and/or confinement radius ( Figure 1D, E, Video 2), while pLAT à N-WASP had a tendency to align with actin filaments and showed a decrease in the mobile fraction and confinement radius ( Figure 1D, E, Video 2). pLAT à Nck exhibited an intermediate behavior, with a slight increase in the fraction of mobile condensates and their apparent diffusion coefficient, but at the same time a decrease in the confinement radius of confined and immobile condensates ( Figure 1D, E, Video 2). pLAT à Nck also tended to align with actin filaments, although to a lesser degree than pLAT à N-WASP ( Figure 1B, C). Lastly, in the presence of active actomyosin networks, condensates of all compositions exhibited an overall increase in mobility (larger mobile fraction, apparent diffusion coefficient, and/or confinement radius). The increase for pLAT à Sos1 and pLAT à pSLP76 was subtle, larger for pLAT à Nck, and largest for pLAT à N-WASP. The influence of actomyosin on pLATà N-WASP condensates was generally in the opposite direction of the influence of actin alone ( Figure 1D, E, Video 3). These data suggest that condensates 8 containing Nck or Nck and N-WASP, which wet filaments and show a large differential in behavior between actin alone and actomyosin conditions, can be viewed distinctly from condensates containing Sos1 or Sos1 and pSLP76, which do not wet filaments and show small differences between the two types of actin networks.

LAT condensates in vitro containing Nck and N-WASP move with contracting actomyosin networks with high fidelity
To delineate the effect of composition on the ability of LAT condensates to move with actin filaments, we devised an in vitro system where the actin filaments moved in a directional manner. This system enabled us to clearly distinguish between condensate that move with the actin filaments (because they would also exhibit directional movement) and those that do not. compared with adjacent, non-imaged regions of the SLB. These actomyosin contraction experiments were performed with pLAT à Sos1 and pLAT à N-WASP condensates. In the beginning of such experiments, we found that pLAT à Sos1 condensates were randomly distributed on the membrane while pLAT à N-WASP condensates were aligned along the actin filaments as observed above (Figure 2A). In both cases the filament network started to contract immediately upon myosin II addition and formed stable asters within 2 minutes (Video 4). As shown in Figure 2A, at the end of the contraction, most of the pLAT à Sos1 condensates remained scattered across the SLB, while virtually all of the pLAT à N-WASP condensates had moved with the actin into asters. To quantify these behaviors, we examined the speed and direction of condensate movement and actin movement during actomyosin network contraction using Spatio-Temporal Image Correlation Spectroscopy (STICS) (Ashdown et al., 2014) ( Figure   2B). We found that the speed of pLAT à N-WASP condensates correlated well with the speed of actin, while the speed of pLAT à Sos1 condensates did not ( Figure 2C). Additionally, the distribution of angles between the vectors of pLAT à N-WASP condensate movement and proximal actin movement showed clear preference for smaller angles, indicating a high degree of co-movement. In contrast, the angle distribution for pLAT à Sos1 showed only a slight preference for smaller angles, which was marginally significant ( Figure 2D). Together, the steady state ( Figure 1) and contraction ( Figure 2) experiments and analyses show that LAT condensates are influenced by actin network dynamics in a composition-dependent fashion.
Condensates containing Nck or N-WASP bind to and move with actomyosin filaments. In contrast, condensates lacking these proteins do not bind filaments appreciably, and are likely moved by non-specific steric contacts. Thus, Nck and N-WASP function as a molecular clutch between LAT condensates and actin.

Basic regions of Nck and N-WASP couple LAT condensates to actin filaments
The data above suggest that Nck and N-WASP mediate binding of LAT condensates to actin filaments. To test this, we quantified the recruitment of preformed actin filaments to SLBs by LAT condensates of different compositions in the absence of eABD. As shown in Figure 3A, only condensates containing Nck or Nck and N-WASP recruited substantial amounts of actin filaments. In both cases, condensates were deformed and elongated along filaments, as observed above.
We next performed co-sedimentation assays to determine whether Nck or N-WASP could bind actin filaments in solution or whether efficient binding required the proteins to be arrayed on a two-dimensional surface. Nck did not appreciably co-sediment with actin filaments. Consistent with previous reports, N-WASP did bind filaments (Co et al., 2007), although not to the same degree as α-actinin, a high-affinity actin filament-binding protein ( (Banjade et al., 2015). Of these seven elements, two contain dense basic patches, one contains dense acidic patches, and the remainder are relatively free of dense charge patches. As detailed in Figure 3 -figure supplement 2, Nck fragments containing an excess of basic elements recruited actin to the membrane, where greater excess resulted in more efficient recruitment, while neutral or acidic fragments did not.
Similarly, mutating one of the basic elements (the linker between the first and second SH3 domains, L1) to neutralize it (Nck Neutral ) or to make it acidic (Nck Acidic ) greatly impaired actin recruitment, while making it more basic (Nck Basic ) enhanced actin recruitment ( Figure 3B, Figure   3 -figure supplement 3). These data indicate that basic regions of Nck likely contribute to binding of LAT condensates to actin filaments.
Like Nck, N-WASP also has a central basic region (amino acid residues 186-200). We thus asked whether this region and/or the two C-terminal WH2 motifs contribute to the coupling of LAT condensates to actin filaments (Bieling et al., 2018;Co et al., 2007). We generated pLAT à N-WASP condensates with N-WASP fragments consisting of the basic-proline elements (BP), basic-proline + VCA (BPVCA) elements, and basic-proline + VCA mut (BPVCA mut ), which contains mutations to the WH2 motifs in the VCA region that impair filament binding (Co et al., 2007). All three types of condensates strongly recruited actin, indicating that WH2-actin interactions are not needed for actin filament recruitment in the context of LAT condensates To test whether these basic region-mediated interactions are necessary to couple LAT condensate movement to actin movement ( Figure 2), we performed actin contraction assays with condensates containing N-WASP Basic or N-WASP Neutral . Before myosin II addition, pLAT -> N-WASP Basic condensates aligned almost perfectly with actin filaments, while pLAT -> N-WASP Neutral condensates only partially aligned with actin filaments, consistent with the notion that the basic region mediates binding to actin filaments ( Figure 3D). After actin network contraction, pLAT à N-WASP Basic condensates colocalized with actin asters to a similar degree as pLAT à N-WASP WT , while pLAT à N-WASP Neutral did not ( Figure 3D vs. Figure 2A, Video 5 vs. Video 4). STICS analysis revealed that the correlation of pLAT à N-WASP Basic movement with local actin movement was slightly better than that of pLAT à N-WASP WT , while the correlation of pLAT à N-WASP Neutral was worse ( Figure 3E, F). The remaining correlation for pLAT à N-WASP Neutral is most likely due to the presence of Nck, which also contributes to condensate binding to actin. Together these data demonstrate that regions of Nck and N-WASP that contain dense basic patches can mediate the clutch-like behaviors of the proteins by directly interacting with actin filaments proportionally to the degree of positive charge, and that these interactions are necessary for LAT condensates to faithfully move with actin.
The composition of LAT condensates changes as they traverse the IS

12
In Jurkat T cells activated by OKT3, an antibody that binds to the CD3ε subunit of the TCR, and ICAM-1 bound to an SLB, LAT condensates that form at the periphery of the IS move to its center over ~5 minutes as the cell-SLB contact matures. To investigate whether the composition-dependent interactions observed in our biochemical data have consequences for LAT condensate behavior in cells, we examined the composition of LAT condensates as they moved in the plane of the plasma membrane during activation of live Jurkat T cells. We used Jurkat T cells because they retain many features of primary T cells relevant to movement and signaling from LAT condensates, but are easier to manipulate and analyze. In both cell types, LAT condensate formation is well-documented (Balagopalan et al., 2013;Lin et al., 1999;Su et al., 2016;Yokosuka et al., 2005), condensate movement across the IS is correlated with actin flow (DeMond et al., 2008;Kaizuka et al., 2007;Murugesan et al., 2016;Yi et al., 2012), and proximal biochemical signaling from LAT through SLP-76 is similar (Bartelt et al., 2009).
However, the IS between Jurkat T cells and supported lipid bilayers is larger than that of primary T cells (Murugesan et al., 2016), and LAT condensates do not initiate actin polymerization to the same degree in Jurkat T cells as in primary T cells, which allows us to analyze the ability of condensates to couple to dynamic actin networks without accounting for their own selfgenerated polymerized actin (Kumari et al., 2015).
We co-expressed LAT-mCitrine with Grb2-mCherry or LAT-mCherry with Nck-sfGFP in Jurkat T cells. These cells bound to SLBs coated with mobile ICAM-1 and OKT3 producing an IS mimic with the SLB. Of note, in contrast to Jurkat T cells adhered to immobile substrates, which extend long lamellipodia (Babich et al., 2012), Jurkat T cells adhering to fluid SLBs here extend short lamellipodia at the IS periphery. We used total internal reflection fluorescence (TIRF) microscopy to capture images of activated cells every 5 seconds for up to 5 minutes. We then automatically detected and tracked LAT condensates from their formation at the periphery of the IS to their coalescence with the central supramolecular activation complex (cSMAC) at the synapse center (Jaqaman et al., 2008)  condensates. In order to follow the LAT condensates neutrally, i.e. while blind to their Grb2 or Nck content, we detected and tracked them based on the LAT channel, and then read out the Grb2 or Nck intensities at the LAT condensate locations ("master/slave" channel analysis, as in (Loerke et al., 2011)). In addition to affording neutrality, this scheme ensured accurate LAT condensate detection and tracking because of the overall better signal in the LAT channel, especially when compared to the Nck channel, which tended to have high background due to soluble Nck molecules not bound to LAT condensates. Furthermore, to overcome the stochasticity and noise inherent to live-cell image data, we aligned tracks (in space or time, as described below) and averaged them to uncover underlying overall trends. For meaningful alignment and averaging, we filtered tracks by their duration, extent of directed movement, initial and final positions (to ensure that they traversed a sufficient radial distance across the IS), and initial Grb2 or Nck intensity to ensure that changes in intensity could be measured accurately (see Materials and Methods for more details).
This analysis revealed that Grb2 colocalized with LAT condensates at the edge of the synapse ( Figure 4A, Video 7) and its fluorescence intensity in the condensates was maintained throughout the trajectory to the center of the synapse ( Figure 4B, C). In contrast, while Nck also colocalized with LAT condensates at the edge of the synapse ( Figure 4D, Video 8), its fluorescence diminished relative to LAT during the trajectory. This intensity decrease began at 0.6-0.7 of the distance from the center of the synapse to the synapse edge (referred to hereafter as "normalized radial position," equal to zero at the synapse center and one at the synapse edge; see Materials and Methods) ( Figure 4C, E). The emergence of this pattern from averaging 125 tracks from 25 cells suggests that spatial position is a key determinant of Nck residence in LAT condensates. In contrast, when the condensates were aligned according to . This suggests that in this experimental setting, position plays a more instructive role than time in determining the residence of Nck in condensates (and presumably the residence of WASP, which is recruited to LAT condensates via Nck). However, since time and space are coupled, our data do not rule out a role for time in this process, as has previously been observed in experimental conditions where LAT condensates were immobilized (Barda-Saad et al., 2005).

Nck dissipation coincides in space with the switch in actin architecture from dendritic network to concentric arcs
Translocation of LAT condensates from the synapse periphery to the center of the IS is driven by motion of the actin cytoskeleton (DeMond et al., 2008;Kaizuka et al., 2007;Mossman et al., 2005;Yu et al., 2010). Recent work has shown that two actin networks are generated at the IS in activated T cells (Murugesan et al., 2016;Yi et al., 2012). In the outer ~1/3 of the synapse, the Arp2/3 complex generates a dendritic actin meshwork, where the filaments are on average directed radially, perpendicular to the synapse edge. In the medial region closer to the cSMAC, this meshwork is largely replaced by formin-generated concentric actin arcs that are directed parallel to the synapse edge ( Figure 4 -figure supplement 4) (Hammer 3rd and Burkhardt, 2013;Yi et al., 2012). Both filament networks move through the action of myosin motors as the cell-cell conjugate matures; however, the nature of this movement is different in the two cases.
The outer dendritic network moves in a direction perpendicular to the edge of the synapse in a process termed retrograde flow (DeMond et al., 2008;Kaizuka et al., 2007;Mossman et al., 2005;Yu et al., 2010), analogous to actin flow observed at the leading edge of migrating cells (Ponti et al., 2005(Ponti et al., , 2004. In contrast, the inner concentric arcs sweep toward the center of the synapse in a telescoping manner and appear to have components of motion both perpendicular and parallel to the synapse edge (Murugesan et al., 2016).
The normalized radial position at which LAT condensates started to lose Nck is similar to the position where the actin network has been reported to change from dendritic architecture to arc architecture (Hammer 3rd and Burkhardt, 2013;Murugesan et al., 2016;Yi et al., 2012). To corroborate this in our cells, we incubated Jurkat T cells with the dye SiR-Actin, which binds to actin filaments in a defined orientation (perpendicular to the filament orientation; Figure 4 figure supplement 5) (Nordenfelt et al., 2017). This dye enabled us to use instantaneous polarization TIRF microscopy (Mehta et al., 2016) to evaluate the orientation of actin filaments at the IS. We found that SiR-actin at concentrations higher than 50nM blocked actin flow at the IS, but speckle labeling of actin networks with 10 nM SiR-actin produced normal actin flow toward the center of the IS. Since the actin networks were stained as sparsely distributed speckles of SiR-actin, our analysis procedure does not involve (or require) identification of individual filaments in the images and then measuring their orientations. Rather, by observing the fluorescence polarization orientation of single speckles randomly bound to actin filaments, we can build a spatial map of filament orientations across the IS. We found that actin filaments in the outer 30% of the synapse were generally oriented perpendicular to the synapse edge, while those closer to the center of the synapse were parallel to the synapse edge ( Figure 4F), in good agreement with earlier super-resolution work (Murugesan et al., 2016). The change in actin filament polarization occurred at 0.6-0.7 of the distance from the center of the synapse, which correlates well with the position at which Nck dissipated from LAT condensates. This position is unrelated to the overall three-dimensional geometry of the cell, indicating that actin architecture at the IS, as observed via TIRF microscopy, is not determined by the position of the cell body above the membrane surface ( Figure 4 -figure supplement 6).

Constitutive engagement of LAT condensates with actin leads to their aberrant movement across the IS
Our combined biochemical and cellular data thus far indicate that Nck and WASP mediate LAT condensate engagement with actin, and that LAT condensates lose these proteins as they move from the dendritic actin meshwork in the outer part of the synapse to the contractile arcs network closer to the synapse center. The biochemical data suggest that this change in composition should allow LAT condensates to interact differently with the two actin networks.
We hypothesized that this switch in interaction might be necessary for the proper radial movement of LAT condensates at the IS, given the different orientation (perpendicular vs.
parallel to the synapse edge) and movement (retrograde flow vs. telescoping motion) of filaments in the two actin networks at the IS. To test this hypothesis, we altered the adhesion of LAT condensates to the actin filament network by fusing Grb2, which remains in the condensates throughout their trajectories ( Figure 4A-C), with the doubled basic region of N-WASP (Grb2 Basic ). In biochemical assays, this generated LAT condensates that bound actin filaments in the absence of Nck or WASP. The pLAT-Grb2 basic complex recruited actin filaments to SLBs, while pLAT alone or the pLAT-Grb2 complex did not ( Figure 5A). In actomyosin contraction assays, condensates of pLAT/Grb2 Basic /Sos1 initially wet filaments and then localized to actin asters after myosin II-induced contraction to a greater degree than pLAT/Grb2/Sos1, although to a lesser degree than pLAT à N-WASP ( Figure 5B, Video 9 vs. Video 4). Similarly, during actomyosin network contraction, the movement of pLAT/Grb2 Basic /Sos1 condensates was correlated more strongly with actin movement than condensates containing Grb2 ( Figure 5 -figure supplement 1), but less strongly than pLAT à N-WASP condensates ( Figure 2C, D). Together, these data demonstrate that the double basic motif of N-WASP, when added to Grb2, can act as a molecular clutch coupling LAT condensates to actin.
We next asked whether expression of Grb2 Basic in Jurkat T cells would perturb the radial movement of LAT condensates due to their constitutively engaged clutch, including in the medial region of the IS where they encounter actin arcs. This was quantified as deviation from a straight path between the start of persistent inward radial movement and coalescence with the cSMAC (see Supplemental Methods for more details). For this, cells expressing Grb2 Basic -mCherry were activated on SLBs as above. Similar to cells expressing Grb2-mCherry, LAT condensates that formed at the periphery of the IS retained Grb2 Basic -mCherry throughout their trajectories to the cSMAC ( Figure 5C, D). However, evaluation of condensate trajectories revealed that they deviated from a straight path significantly more than condensates in cells expressing Grb2-mCherry ( Figure 5E, F, Video 10 vs. Video 7). This behavior is consistent with abnormally high adhesion of condensates containing Grb2 Basic to actin filaments, even after Nck has presumably dissipated, leading to trajectories that reflected more the telescoping, circular component of the contractile actin arc motion.

Formin activity is necessary for LAT condensate composition change
Finally, we asked whether the transition from the dendritic actin architecture to the contractile arc architecture might play a role in changing the composition of LAT condensates at this location in the IS. Previous data showed that the contractile arcs are generated by the formin mDia1 and could be eliminated by cell treatment with the formin inhibitor, SMIFH2 (Murugesan et al., 2016). We found that in contrast to control cells (treated with DMSO), where Nck dissipated normally from LAT condensates ( Figure 6A and B, Figure 6 -figure supplement 1, Video 11), cells treated with SMIFH2 for five minutes prior to imaging displayed LAT condensates with virtually constant Nck intensities throughout their trajectories from the periphery to the cSMAC ( Figure 6A and C, Figure 6 -figure supplement 1, Video 12). Thus, the activity of formin proteins, and/or perhaps the actin arcs that they generate, act to alter the composition of LAT condensates, likely altering their downstream signaling activities in the central region of the IS. We note that the SMIFH2 data further support the notion that in unperturbed cells, space, rather than time, is the key determinant of Nck residence in condensates (assuming that formin does not also create a temporal signal). Our combined data suggest that the two actin networks in activated Jurkat T cells not only spatially organize the immunological synapse by moving LAT condensates, but may also contribute to creation of specific signaling zones.

Discussion
Compositional changes of biomolecular condensates in response to signals have been well documented (Chen et al., 2008;Dellaire et al., 2006;Markmiller et al., 2018;Salsman et al., 2017;Youn et al., 2018 repeated non-specific steric interactions. This is consistent with previous observations that in the actin arcs region of the IS, LAT condensates are repeatedly hit by arcs that move them briefly but then release (Murugesan et al., 2016). The circular movement of the telescoping actin arcs is randomly directed clockwise and counterclockwise, and repeated hits by arcs moving oppositely produce no net circular motion on the condensates. But the radial component is consistently directed toward the center of the IS, and thus repeated hits add constructively to produce a net movement in a radial direction. Thus, LAT condensates continue to move linearly toward the center of the IS in the arc region. If the condensates were to adhere tightly to actin in the arcs region (i.e. if they contained Nck and WASP there), they would no longer undergo repeated hits by arcs moving in opposite directions, and the circumferential force would not average to zero. In the simplest case, they would attach to the first arc they encountered and move circumferentially with it. Such effects likely account for the aberrant movement of LAT condensates containing the Grb2 mutant artificially equipped with a basic clutch, which should produce inappropriately strong adhesion to the telescoping actin arcs.
While we have examined Jurkat T cells here, existing data suggest that the mechanisms we have discovered for LAT condensate movement are likely similar to those in primary T cells.
The IS formed by both Jurkat T cells and primary T cells is composed of a dendriticallybranched actin network at the edge of the IS, followed immediately by a concentric actomyosin cable network near the center of the IS (Murugesan et al., 2016). Thus, condensates must be moved across two distinct actin networks in Jurkat and primary T cells. In primary T cells, condensate-associated actin polymerization localizes mostly (although not entirely) in the outer region of the IS and dissipates in the region adjacent to the branched actin network, where ICAM-1 localizes (Kumari et al., 2015), which would be consistent with loss of Nck and WASP toward the center of the IS. It is tempting to speculate that the engagement of the clutch mechanism has mechano-signaling consequences, since T cell receptor signaling appears to be mechanically gated (Chen and Zhu, 2013;Hui et al., 2014). Similarly, PLC-γ1 activation by WASP-promoted actin polymerization appears to localize mostly in the dSMAC where we observe strong Nck co-localization with LAT (Kumari et al., 2015). One difference between the two cell types is that WASP-promoted actin polymerization is much weaker in Jurkat T cells than in primary T cells (Kumari et al., 2015). In primary cells, this actin assembly may also play a role in the movement of condensates from the cell edge to the center, in addition to myosindriven movement of the cortical actin. Future work addressing the modes of movement, and the precise signals that dictate compositional change, will elucidate the mechanisms by which LAT condensates move across the IS in primary T cells.
LAT condensates represent one particular type of biomolecular condensate. It is generally thought that the functions of condensates are intimately connected to their compositions, and 21 that changes in composition could cause changes in function (Banani et al., 2016). Our data here demonstrate that when Nck and N-WASP are arrayed on membranes they can bind actin filaments efficiently, even though both bind filament sides only weakly in solution. This adhesion enables condensates containing the proteins to be moved over long distances in response to actomyosin contraction. Adhesion is lost when Nck and N-WASP depart. Thus, the composition of LAT condensates plays an important role in their coupling to actin and their mode of movement at the IS. These behaviors of the LAT system are produced by generalizable features of membrane-associated condensates -their high density and composition based on regulatable interactions. Analogous behaviors are likely to be widely observed as the biochemical and cellular activities of other condensates are explored.

LAT Purification and Modification
glycerol, and eluted with 500 mM imidazole (pH 8.0), 150 mM NaCl, 5 mM βME, 0.01% NP-40, and 10% glycerol. The MBP tag and His 6 tag were removed using TEV protease treatment for 16 hrs at 4°C. Cleaved protein was applied to a Source 15 Q anion exchange column and eluted with a gradient of 200 mMà300 mM NaCl in 20 mM HEPES (pH 7.0) and 2 mM DTT followed by size exclusion chromatography using a Superdex 200 prepgrade column (GE Healthcare) in 25 mM HEPES (pH 7.5), 150 mM NaCl, 1 mM MgCl 2 , and 1 mM DTT. LAT was concentrated using Amicon Ultra Centrifugal Filter units (Millipore) to >400 μM, mixed with 25 mM HEPES (pH 7.5), 150 mM NaCl, 15 mM ATP, 20 mM MgCl 2 , 2 mM DTT, and active GST-ZAP70 (SignalChem), and incubated for 24 hrs at 30°C. Phosphorylated LAT (pLAT) was resolved on a Mono Q anion exchange column using a shallow 250 mM à 320 mM NaCl gradient in 25 mM HEPES (pH 7.5), 1 mM MgCl 2 , and 2 mM βME to separate differentially phosphorylated species of LAT. Complete LAT phosphorylation was confirmed by mass spectrometry. pLAT was then exchanged into 25 mM HEPES (pH 7.0), 150 mM NaCl, and 1 mM EDTA (pH 8.0) using a HiTrap Deslating Column (GE Healthcare). C 5 -maleimide Alexa488 was added in excess to pLAT in reducing agent-free buffer and incubated for 16 hrs at 4°C.
Following the incubation 5 mM βME was added to the labeling solution to quench the reaction.

Steady-State Reconstitution Assays
These methods are adapted from previously published methods (Köster et al., 2016;Su et al., 2017Su et al., , 2016 Table S2 for more information.

Actomyosin Contraction Assays
These methods are adapted from previously published methods (Köster et al., 2016;Su et al., 2017Su et al., , 2016 (Nordenfelt et al., 2017). Briefly, the first ten frames following IS formation from each time-lapse image set were analyzed using custom code developed in MATLAB 2014a. Calculated polarizations for each detected SiR-Actin speckle were visualized on time-lapse image panels using FIJI and plotted on a radial line from the synapse center to the synapse edge and determined to be within 45° of perpendicular or parallel form the synapse edge.
Confocal Microscopy of Activated Jurkat T cells.
Confocal images were captured using a Yokogawa spinning disk (Biovision) mounted on a Leica DMI6000 microscope base equipped with a Hamamatsu ImagEMX2 EM-CCD camera with a 100X 1.49 NA objective. Images were acquired using MetaMorph.software. SLBs were prepared for cellular activation as described above. Jurkat T cells expressing LAT-mCherry, Nck-sfGFP, and LifeAct-BFP were activated on the SLB for 5 minutes to allow the IS to form.
Confocal slices were then captured with a 0.25 μm step-size. 3-dimensional images were reconstructed using Matlab and the position of the dense actin ring (LifeAct-BFP) and membrane (as indicated by LAT-mCherry fluorescence) were measured and analyzed for spatial orientation using Matlab.
Data Analysis and Display

Drift Correction for In Vitro Movies
Due to imaging multiple time points at multiple stage positions for all in vitro reconstitution assay imaging, movies were subject to drift artifacts. To correct for drift, we aligned the frames in the pLAT channel by the maximum pixel-level cross correlation between adjacent frames. We then applied the shift to all channels to ensure identical corrections.
There were no drift artifacts from cellular imaging data because a single stage position was used.

pLAT Condensates in vitro within No Actin / Actin / Actomyosin Networks at Steady State 2a. Detection of pLAT Condensates
To detect pLAT condensates, we first pre-processed the pLAT condensate images by (i) subtracting inhomogeneous background, where the background image was estimated by filtering the pLAT image with a large Gaussian kernel (σ = 10 pixels), and (ii) suppressing noise by filtering the background subtracted image with a small Gaussian kernel (σ = 1). Next we detected pLAT condensates using a combination of local maxima detection (Jaqaman et al., 2008) to handle diffraction-limited condensates, which tended to be dim, and intensity-based segmentation (Otsu threshold) to handle larger condensates, which tended to be brighter.
Applying these two algorithms resulted in three detection scenarios: (1) A segmented region containing one local maximum within it, taken to represent one condensate.
(2) A local maximum not enclosed within a segmented region, representing a dim, diffraction-limited condensate. In this case, a circular area approximating the two-dimensional point spread function (PSF σ = 74 nm = 0.46 pixels, => circle radius = 3σ = 222 nm = 1.4 pixels) centered at the local maximum was taken in lieu of a segmented condensate area.
(3) A segmented region containing multiple local maxima. This scenario could arise from overlapping nearby condensates or from fluctuating intensity within one large condensate (possibly due to incomplete mixing following the fusion of two or more individual condensates). To distinguish between these two cases, we compared the intensity variation along the line connecting each pair of local maxima in a segmented region to the intensity variation between the center and the edge of isolated condensates (scenario 1, i.e. one local maximum within one condensate).
Specifically, for each pair of local maxima in a segmented region, we averaged their peak intensities and then calculated the ratio of this average to the minimum intensity along the line between them. Similarly, for isolated condensates (scenario 1), we calculated the ratio of their peak intensity to the minimum intensity at the edge (Figure 1 -figure supplement 1). This provided us with a reference distribution of peak/minimum ratios for truly separate condensates.
In particular, we took the 1 st percentile of this distribution as a threshold to distinguish between pairs of local maxima with a segmented region belonging to separate condensates (pair peak/minimum ratio ≥ threshold) and pairs of local maxima belonging to the same condensate (pair peak/minimum ratio < threshold). If a pair of local maxima was deemed to belong to the same condensate, the local maximum with lower peak intensity was discarded. After eliminating superfluous local maxima with this procedure, we finalized the condensate segmentation and center estimation by using watershed on the originally segmented region with the remaining local maxima as seeds for the watershed algorithm (Figure 1 -figure supplement   1).

2b. Tracking of pLAT Condensates
To track the detected pLAT condensates, we employed our previously developed multipleparticle tracking software "u-track" (Jaqaman et al., 2008), using a search radius of 5 pixels, a gap closing time window of 3 frames, and with the possibility of merging and splitting (all other parameters had default values). The search radius and gap closing time window were optimized by inspecting the distributions of frame-to-frame displacements and gap durations output by the software, and by visual inspection of the tracked condensates.

2c. Motion Analysis of pLAT Condensates
To characterize the movement of pLAT condensates, we used previously developed moment scaling spectrum (MSS) analysis of the condensate tracks (Ewers et al., 2005;Ferrari et al., 2001;Jaqaman et al., 2011). Apart from employing new MSS slope thresholds to distinguish between immobile and confined tracks (Vega et al., 2018), all analysis procedures and parameters were as described in (Jaqaman et al., 2011).

2d. Analysis of Actin Enrichment at pLAT Condensates
To quantify the enrichment of actin at pLAT condensates ( Figure 1C), we employed our previously developed point-to-continuum colocalization analysis algorithm (Githaka et al., 2016).
Briefly, actin enrichment in each movie was defined as the ratio of actin intensity within pLAT condensates to the actin intensity outside condensates, averaged over all condensates, at the last frame of each movie.

pLAT Condensates In Vitro within Contractile Actomyosin Networks 3a. STICS Analysis of pLAT and Actomyosin Movement
To capture the overall movement of the actomyosin network and pLAT condensates, we first applied SpatioTemporal Image Correlation Spectroscopy (STICS) analysis to each channel separately (Ashdown et al., 2014). STICS was preferred over particle tracking to analyze the pLAT channel in this assay for two reasons: 1) Experimental conditions (described above) resulted in many pLAT à N-WASP condensates strongly wetting actin filaments (when

3b. Analysis of pLAT and Actomyosin Co-movement
After acquiring STICS movement vector fields for the two channels (over the contraction time interval), we compared the vector magnitudes (i.e. speeds) and angles at each image subregion between the two channels. The angle distribution (e.g. Figure 2D; taken from the histogram function and plotted using the stairs function) and speed comparison (e.g. Figure 2C; plotted using the MATLAB function histogram2) for any condition were comprised of all the measurements from all sub-regions of all movies representing that condition. Angle distributions were compared between conditions (or between a condition and its randomized control) via a Kolmogorov-Smirnov (KS) test. For this, due to the hyper-sensitivity of the KS test when the distributions comprise many data points (roughly >1000 data points), we performed the KS test on pairs of subsamples from the distributions (down to 500 data points each), repeated 100 times, and the average p-value of the 100 repeats was reported as the KS test p-value (e.g. in Figure 2D).

Live Cell Data 4a. Synapse and cSMAC Segmentation
To identify the location of LAT condensates relative to the synapse edge and cSMAC center, we first segmented the synapse and the cSMAC. Both were done in a semi-automated manner.
For synapse segmentation, the actin channel (LifeAct -BFP) was used when available for a condition, as this channel provided the clearest synapse edge for segmentation. In conditions where actin was not labeled, the LAT channel was used. Images were first smoothed using a Gaussian kernel (σ = 2 pixels), and then an Otsu threshold was applied to separate the image into 2 levels, i.e. synapse and background (using the MATLAB function multithresh, which generalizes the Otsu method to determine thresholds for a multimodal distribution). We retained only the synapse at the center of the image for further analysis, identified as the largest thresholded object in the image.
For cSMAC segmentation, the LAT channel was always used. In this case, multithresh was tasked to determine two thresholds that would separate the LAT image into three levels. The cSMAC was taken as the largest segmented area at the highest intensity level within the segmented synapse. For early frames in which the cSMAC had not yet formed, instead of segmenting the cSMAC we used the point that would eventually become the cSMAC center as an alternative reference. Specifically, we applied the above process but on the average of all the time-lapse frames, and through this the center of the eventual cSMAC was determined and used in those early frames (area = 0 in this case since cSMAC had not formed yet).
We then visually inspected all segmentation results (synapse and cSMAC) and in some cases manually refined the segmentation using in-house software.

4b. Detection and Tracking of LAT Condensates
Due to the majority of condensates being diffraction limited and a lower SNR in our cellular data, thresholding as described above in 2a for in vitro condensates was not appropriate for cellular analysis, as it lacked the sensitivity to detect individual condensates in cells. Instead, we detected pLAT condensates solely using local maxima detection (Jaqaman et al., 2008).
After detection, we tracked the LAT condensates in the same way and with the same parameters described in section 2b for in vitro condensates.

4c. Defining a Normalized Radial Position Between Synapse Edge and cSMAC Center
Because synapse and cSMAC size differed between cells, and additionally synapses were not circular, we sought a unitless, normalized measure of position that allowed us to pool measurements between cells, and even from different parts of the same cell. To this end, for any point in the synapse, we drew a straight line from the cSMAC center to the synapse edge going through that point, and then defined the point's normalized radial position as the ratio of the distance between the point and the cSMAC center to the total line length. With this, the normalized radial position ranged from zero at the cSMAC center to one at the synapse edge.
All LAT condensate and actin trends were then measured vs. this normalized radial position (e.g. Figure 4C).

4d. LAT Condensate Composition Analysis
To analyze the composition of LAT condensates as they moved from the synapse edge toward the cSMAC, we selected condensates based on their track duration, track geometry, track start and end position, and initial Nck/Grb2 content, as explained next in detail: (1) Track duration: Only tracks lasting a minimum of 5 frames were used for any analysis, to obtain enough information per track.
(2) Track geometry: We reasoned that LAT condensates moving toward the cSMAC should be overall linear (asymmetric). Therefore, for this analysis we selected tracks that were approximately linear, as assessed by measuring the degree of anisotropy of the scatter of condensate positions along the track (Huet et al., 2006;Jaqaman et al., 2008).
(3) Track start and end position: In addition to being approximately linear, tracks used for composition analysis were filtered by their start and end positions. To this end, we used the rise and fall of actin intensity as one traversed the synapse from the edge to the cSMAC to define a position threshold (described next), such that LAT condensate tracks used for composition analysis had to start between the synapse edge and this threshold, and had to end between this threshold and the cSMAC center.
To define this position threshold, we took for every cell at each frame eight actin intensity profiles from the synapse edge to the cSMAC center, using straight lines with a (4) Initial Nck/Grb2 content: As explained in the next paragraph, for composition analysis the amount of Nck/Grb2 in a LAT condensate at any time point was normalized by its amount when the condensate first appeared (specifically the average of the first three frames to account for fluctuations). To exclude condensates that had low amounts of Nck or Grb2 to begin with, only condensates that contained an average Nck or Grb2 protein intensity (as defined next) in their first three time points greater than the average standard deviation of the background intensity in the first three time points were included in the analysis (on average 88% of the tracks surviving conditions 1-3 above).
To quantify protein (i.e. Nck, Grb2, or LAT) content in a condensate, we subtracted local background from the protein intensity inside the condensate and then took the average background subtracted intensity as a measure of protein content. To estimate local background, we determined which condensates were in each other's proximity (referred to as "condensate aggregates") and then calculated the average and standard deviation of intensity in a 2 pixel thick perimeter around each condensate aggregate (thus all condensates within one aggregate had the same local background level). For composition analysis over the lifetime of each track, the protein content at each of its time points was normalized by its average protein content at its first three time points. The normalized protein content was then pooled for all condensates based on their frame-by-frame normalized radial position, specifically 0.9-1 (closest to the synapse edge), 0.8-0.9, 0.7-0.8, etc. The ratio of pooled normalized protein content (either Grb2 : LAT or Nck : LAT) was plotted as a minimal boxplot showing only the median and notches, indicating the 95% confidence interval of each median (e.g. Figure   4C).

4e. Measuring Deviation from Straight Path for Condensate Tracks
In order to measure whether condensates retaining a molecular clutch via Grb2 Basic displayed aberrant movement as they traversed the two actin networks, we measured the deviation of each condensate track from a straight path towards the cSMAC. Before measuring track deviation from a straight path, it was necessary to remove the following two parts of each To identify the initial part of a track, i.e. condensate movement along the synapse edge rather than toward the cSMAC, we measured both the frame-to-frame change in distance between the condensate and the cSMAC center and the frame-to-frame change in angle between consecutive displacements taken by the condensate. The rationale was that before a condensate began to move towards the cSMAC, the change in distance should be low, while the change in angle should be high. On the other hand, once a condensate began to move towards the center, the change in distance should be high, while the change in angle should be low. Therefore, to detect the transition between moving along the synapse edge and moving toward the cSMAC, we took the ratio of the change in distance to change in angle at every track After the initial and final track parts were removed, the remaining track segment was transformed such that the beginning and end were set to zero on the y-axis. Thus any non-zero y-position along the track could be directly measured as a deviation from a straight path.
Additionally, tracks were flipped if necessary so that the majority of their deviations were in the positive direction. The entire deviation distribution from all time points of all tracks was then taken from the histogram function and plotted using the stairs function ( Figure 5F). Example tracks were plotted using the function scatter and colored to represent the deviation value at each position ( Figure 5E).

4f. Measuring Extent of Photobleaching in Live Cell Data
In order to determine whether photobleaching might contribute to measured changes in condensate composition, we measured the change in cell background intensity over time. The mean intensity within the segmented synapse, but outside the segmented cSMAC and detected condensate areas, was taken for each cell for every frame. All mean intensity measurements were normalized by the mean intensity of the first frame for each cell. The ratio of pooled                       independent experiments and 82 condensates from 11 cells expressing Grb2-mCherry and LAT-mCitrine from independent experiments (same cells as in Figure 4C). P-value is for comparing the two distributions via a Kolmogorov-Smirnov test. Only tracks in which the mean Grb2 or Grb2 Basic intensity was greater than 1 standard deviation above background during the first three measurements were used to generate this plot.    Video 7. Grb2 is maintained as LAT condensates move across the IS in activated Jurkat T cells. TIRF microscopy of an activated Jurkat T cell expressing Grb2-mCherry (magenta) and LAT-mCitrine (green) on a SLB coated with ICAM-1 and OKT3 revealed that Grb2 co-localizes with LAT condensates as they move from the edge of the synapse to the cSMAC. Movie shows a 22 µm x 22 µm field of view. The movie is played at 7 fps with frame intervals of 5 seconds.

Video 8. Nck dissipates from LAT condensates as they move across the IS in activated
Jurkat T cells. TIRF microscopy of an activated Jurkat T cell expressing LAT-mCherry (magenta) and Nck-sfEGFP (green) on a SLB coated with ICAM-1 and OKT3 revealed that Nck dissipates from LAT condensates as they move from the edge of the synapse to the cSMAC.
Movie shows a 24 µm x 24 µm field of view. The movie is played at 5 fps with frame intervals of 5 seconds.

72
Video 9. Reconstitution of movement in LAT condensates containing Grb2 Basic in contracting actomyosin networks. TIRF microscopy revealed actin filament and LAT condensate movement on supported lipid bilayers in a contracting actomyosin network. His 8 -pLAT-Alexa488 (green) was attached to Ni-functionalized SLBs at 500 molecules / µm 2 . Actin filaments (magenta) were attached to the same bilayers via His 10 -ezrin actin binding domains.
Reconstitution was performed in 50 mM KCl. Condensates were formed by adding 125 nM Grb2 Basic and 125 nM Sos1. After formation, condensates bound to and wet actin filaments.
Actin filament contraction was induced by adding 100 nM myosin II. LAT condensates composed of Grb2 Basic moved with contracting actin filaments, although not to the same degree as pLAT à N-WASP WT condensates (Compare with Movie S6). Movie shows a 52 µm x 52 µm field of view. The movie is played at 5 fps with frame intervals of 5 seconds.
Video 10. LAT condensates containing Grb2 Basic deviate tend to deviate from a radial trajectory across the IS. TIRF microscopy of an activated Jurkat T cell expressing Grb2 Basic -mCherry and LAT-mCitrine on a SLBcoated with ICAM-1 and OKT3 revealed that condensates containing Grb2 Basic -mCherry tend to deviate from a radial trajectory as they move across the IS (Compare with Movie S1). Movie shows a 35 µm x 35 µm field of view. The movie is played at 6 fps with frame intervals of 5 seconds.

Video 11. Nck dissipates from LAT condensates as they move across the IS in activated
Jurkat T cells treated with DMSO. TIRF microscopy of an activated Jurkat T cell expressing LAT-mCherry (magenta) and Nck-sfGFP (green) on a SLB coated with ICAM-1 and OKT3 revealed that Nck dissipates from LAT condensates as they move from the edge of the synapse to the cSMAC following treatment with DMSO for 5 minutes prior to activation. Movie shows a 31 µm x 31 µm field of view. The movie is played at 6 fps with frame intervals of 5 seconds.

Video 12. Nck is maintained in LAT condensates as they move across the IS in activated
Jurkat T cells treated with the formin inhibitor SMIFH2. TIRF microscopy of an activated Jurkat T cell expressing LAT-mCherry (magenta) and Nck-sfGFP (green) on a SLB coated with ICAM-1 and OKT3 revealed that Nck is maintained in LAT condensates as they move from the synapse of the cell to the cSMAC following treatment with SMIFH2 for 5 minutes prior to activation. Movie shows a 27 µm x 27 µm field of view. The movie is played at 10 fps with frame intervals of 5 seconds.